Improvements of IP Representation, Fitting and Registration

概要

論文の詳細を見る
Among function-based shape representation techniques, representation in implicit polynomial (IPs) focuses attention in the vision community, because IPs are superior especially in the areas of fast fitting, few parameters, algebraic/geometric invariants, robustness against noise and occlusion, etc. Despite these excellent characteristics, still IP mainly suffers from three issues: difficulty of fine representation for complex objects, difficulty of determining moderate degree for fitting and difficulty of being used for partial object registration. Addressing these issues, in this paper, first a 3D IP-segment representation method is developed even for robustly representing complex objects. Second, an computationally efficient fitting method is proposed for adaptively estimating the IP of moderate degree to the complexity of object shapes. Third, a computationally efficient and robust object registration method using IP gradient field is presented. With these methods, this paper provides the insights and extendible applicabilities into the theory and practice of IP representation.
2009-06-02